This files is a summary of the work made by GMM5 students on data coming from ENVT study.

The data are processed using the following packages:

library(mixOmics)
library(metagenomeSeq)
library(reshape2)

Data description

Data are given in two files included in the directory data :

df_abundances <- read.delim("../data/abondances.csv", sep = ",", 
                            stringsAsFactors = FALSE)
summary(df_abundances[ ,1:15])
##  X.blast_taxonomy   blast_subject      blast_perc_identity
##  Length:406         Length:406         Min.   : 94.18     
##  Class :character   Class :character   1st Qu.: 99.39     
##  Mode  :character   Mode  :character   Median : 99.80     
##                                        Mean   : 99.32     
##                                        3rd Qu.:100.00     
##                                        Max.   :100.00     
##  blast_perc_query_coverage  blast_evalue blast_aln_length
##  Min.   : 98.13            Min.   :0     Min.   :430.0   
##  1st Qu.:100.00            1st Qu.:0     1st Qu.:466.2   
##  Median :100.00            Median :0     Median :480.0   
##  Mean   : 99.96            Mean   :0     Mean   :477.4   
##  3rd Qu.:100.00            3rd Qu.:0     3rd Qu.:490.0   
##  Max.   :100.00            Max.   :0     Max.   :512.0   
##    seed_id          observation_name   observation_sum   
##  Length:406         Length:406         Min.   :   224.0  
##  Class :character   Class :character   1st Qu.:   588.5  
##  Mode  :character   Mode  :character   Median :   979.5  
##                                        Mean   :  9753.7  
##                                        3rd Qu.:  2738.0  
##                                        Max.   :880523.0  
##  NG.10214_EN10A_lib136338_4869_1 NG.10214_EN10B_lib136339_4869_1
##  Min.   :    0.0                 Min.   :    0.0                
##  1st Qu.:    0.0                 1st Qu.:    0.0                
##  Median :    0.0                 Median :    0.0                
##  Mean   :  108.4                 Mean   :  108.4                
##  3rd Qu.:   12.0                 3rd Qu.:   12.0                
##  Max.   :17044.0                 Max.   :17663.0                
##  NG.10214_EN11A_lib136340_4869_1 NG.10214_EN11B_lib136341_4869_1
##  Min.   :    0.0                 Min.   :    0.0                
##  1st Qu.:    0.0                 1st Qu.:    0.0                
##  Median :    1.0                 Median :    1.0                
##  Mean   :  108.4                 Mean   :  108.4                
##  3rd Qu.:    6.0                 3rd Qu.:    6.0                
##  Max.   :24223.0                 Max.   :23739.0                
##  NG.10214_EN15A_lib136342_4869_1 NG.10214_EN15B_lib136343_4869_1
##  Min.   :    0.00                Min.   :    0.00               
##  1st Qu.:    0.00                1st Qu.:    0.00               
##  Median :    3.00                Median :    4.00               
##  Mean   :  108.37                Mean   :  108.37               
##  3rd Qu.:   15.75                3rd Qu.:   15.75               
##  Max.   :28421.00                Max.   :27946.00

The first 9 columns contain information about the different bacteria identified within samples. The following columns contain the two technical replicates (identified by “A” and “B”) of all the farms involved in the study (the farm is identified by a number preceeding the letter “A”/“B”).

Note that some “blast taxonomy” are duplicated:

unique(names(which(table(df_abundances[ ,1]) > 1)))
##  [1] "Bacteria;Actinobacteria;Actinobacteria;Corynebacteriales;Corynebacteriaceae;Corynebacterium 1;Corynebacterium sp."     
##  [2] "Bacteria;Actinobacteria;Actinobacteria;Corynebacteriales;Corynebacteriaceae;Corynebacterium 1;unknown species"         
##  [3] "Bacteria;Actinobacteria;Actinobacteria;Corynebacteriales;Corynebacteriaceae;Corynebacterium;unknown species"           
##  [4] "Bacteria;Actinobacteria;Actinobacteria;Corynebacteriales;Dietziaceae;Dietzia;Dietzia sp."                              
##  [5] "Bacteria;Actinobacteria;Actinobacteria;Corynebacteriales;Nocardiaceae;Rhodococcus;Rhodococcus sp."                     
##  [6] "Bacteria;Actinobacteria;Actinobacteria;Micrococcales;Brevibacteriaceae;Brevibacterium;Brevibacterium antiquum"         
##  [7] "Bacteria;Actinobacteria;Actinobacteria;Micrococcales;Brevibacteriaceae;Brevibacterium;unknown species"                 
##  [8] "Bacteria;Actinobacteria;Actinobacteria;Micrococcales;Intrasporangiaceae;Janibacter;unknown species"                    
##  [9] "Bacteria;Actinobacteria;Actinobacteria;Micrococcales;Microbacteriaceae;Leucobacter;unknown species"                    
## [10] "Bacteria;Actinobacteria;Actinobacteria;Micrococcales;Micrococcaceae;Arthrobacter;Arthrobacter sp."                     
## [11] "Bacteria;Actinobacteria;Actinobacteria;Streptomycetales;Streptomycetaceae;Streptomyces;Streptomyces sp."               
## [12] "Bacteria;Bacteroidetes;Bacteroidia;Bacteroidales;Bacteroidaceae;Bacteroides;unknown species"                           
## [13] "Bacteria;Bacteroidetes;Bacteroidia;Bacteroidales;Porphyromonadaceae;Petrimonas;unknown species"                        
## [14] "Bacteria;Bacteroidetes;Bacteroidia;Bacteroidales;Porphyromonadaceae;Porphyromonas;unknown species"                     
## [15] "Bacteria;Bacteroidetes;Bacteroidia;Bacteroidales;Porphyromonadaceae;Proteiniphilum;unknown species"                    
## [16] "Bacteria;Bacteroidetes;Bacteroidia;Bacteroidales;Prevotellaceae;Alloprevotella;unknown species"                        
## [17] "Bacteria;Bacteroidetes;Bacteroidia;Bacteroidales;Prevotellaceae;Prevotella 1;unknown species"                          
## [18] "Bacteria;Bacteroidetes;Bacteroidia;Bacteroidales;Prevotellaceae;Prevotella 2;unknown species"                          
## [19] "Bacteria;Bacteroidetes;Bacteroidia;Bacteroidales;Prevotellaceae;Prevotella 9;unknown species"                          
## [20] "Bacteria;Bacteroidetes;Bacteroidia;Bacteroidales;Rikenellaceae;Rikenellaceae RC9 gut group;unknown species"            
## [21] "Bacteria;Bacteroidetes;Flavobacteriia;Flavobacteriales;Flavobacteriaceae;Bergeyella;unknown species"                   
## [22] "Bacteria;Bacteroidetes;Flavobacteriia;Flavobacteriales;Flavobacteriaceae;Flavobacterium;unknown species"               
## [23] "Bacteria;Bacteroidetes;Sphingobacteriia;Sphingobacteriales;Sphingobacteriaceae;Olivibacter;unknown species"            
## [24] "Bacteria;Bacteroidetes;Sphingobacteriia;Sphingobacteriales;Sphingobacteriaceae;Pedobacter;unknown species"             
## [25] "Bacteria;Firmicutes;Bacilli;Bacillales;Planococcaceae;Caryophanon;Caryophanon latum"                                   
## [26] "Bacteria;Firmicutes;Bacilli;Bacillales;Staphylococcaceae;Jeotgalicoccus;unknown species"                               
## [27] "Bacteria;Firmicutes;Bacilli;Bacillales;Staphylococcaceae;Staphylococcus;Staphylococcus sp."                            
## [28] "Bacteria;Firmicutes;Bacilli;Lactobacillales;Aerococcaceae;Facklamia;unknown species"                                   
## [29] "Bacteria;Firmicutes;Bacilli;Lactobacillales;Aerococcaceae;unknown genus;unknown species"                               
## [30] "Bacteria;Firmicutes;Bacilli;Lactobacillales;Carnobacteriaceae;Trichococcus;unknown species"                            
## [31] "Bacteria;Firmicutes;Bacilli;Lactobacillales;Carnobacteriaceae;unknown genus;unknown species"                           
## [32] "Bacteria;Firmicutes;Bacilli;Lactobacillales;Enterococcaceae;Enterococcus;unknown species"                              
## [33] "Bacteria;Firmicutes;Bacilli;Lactobacillales;Streptococcaceae;Streptococcus;Streptococcus pluranimalium"                
## [34] "Bacteria;Firmicutes;Bacilli;Lactobacillales;Streptococcaceae;Streptococcus;unknown species"                            
## [35] "Bacteria;Firmicutes;Clostridia;Clostridiales;Clostridiaceae 1;Clostridium sensu stricto 1;unknown species"             
## [36] "Bacteria;Firmicutes;Clostridia;Clostridiales;Family XI;Peptoniphilus;unknown species"                                  
## [37] "Bacteria;Firmicutes;Clostridia;Clostridiales;Lachnospiraceae;Blautia;unknown species"                                  
## [38] "Bacteria;Firmicutes;Clostridia;Clostridiales;Lachnospiraceae;Lachnoclostridium;unknown species"                        
## [39] "Bacteria;Firmicutes;Clostridia;Clostridiales;Lachnospiraceae;Lachnospiraceae NK3A20 group;unknown species"             
## [40] "Bacteria;Firmicutes;Clostridia;Clostridiales;Peptostreptococcaceae;Intestinibacter;unknown species"                    
## [41] "Bacteria;Firmicutes;Clostridia;Clostridiales;Peptostreptococcaceae;Peptoclostridium;unknown species"                   
## [42] "Bacteria;Firmicutes;Clostridia;Clostridiales;Ruminococcaceae;Faecalibacterium;unknown species"                         
## [43] "Bacteria;Firmicutes;Clostridia;Clostridiales;Ruminococcaceae;Ruminococcaceae UCG-005;unknown species"                  
## [44] "Bacteria;Firmicutes;Erysipelotrichia;Erysipelotrichales;Erysipelotrichaceae;Erysipelothrix;unknown species"            
## [45] "Bacteria;Fusobacteria;Fusobacteriia;Fusobacteriales;Leptotrichiaceae;unknown genus;unknown species"                    
## [46] "Bacteria;Proteobacteria;Alphaproteobacteria;Rhizobiales;Hyphomicrobiaceae;Devosia;unknown species"                     
## [47] "Bacteria;Proteobacteria;Alphaproteobacteria;Rhizobiales;Methylobacteriaceae;Methylobacterium;Methylobacterium sp."     
## [48] "Bacteria;Proteobacteria;Alphaproteobacteria;Rhizobiales;Phyllobacteriaceae;Aquamicrobium;unknown species"              
## [49] "Bacteria;Proteobacteria;Alphaproteobacteria;Rhizobiales;Rhizobiaceae;Rhizobium;Agrobacterium tumefaciens"              
## [50] "Bacteria;Proteobacteria;Betaproteobacteria;Burkholderiales;Alcaligenaceae;Alcaligenes;Alcaligenes faecalis"            
## [51] "Bacteria;Proteobacteria;Betaproteobacteria;Burkholderiales;Comamonadaceae;Comamonas;unknown species"                   
## [52] "Bacteria;Proteobacteria;Betaproteobacteria;Burkholderiales;Oxalobacteraceae;Massilia;unknown species"                  
## [53] "Bacteria;Proteobacteria;Gammaproteobacteria;Enterobacteriales;Enterobacteriaceae;Escherichia-Shigella;Escherichia coli"
## [54] "Bacteria;Proteobacteria;Gammaproteobacteria;Pasteurellales;Pasteurellaceae;Mannheimia;Mannheimia sp."                  
## [55] "Bacteria;Proteobacteria;Gammaproteobacteria;Pseudomonadales;Moraxellaceae;Moraxella;Moraxella oblonga"                 
## [56] "Bacteria;Proteobacteria;Gammaproteobacteria;Pseudomonadales;Moraxellaceae;Moraxella;unknown species"                   
## [57] "Bacteria;Proteobacteria;Gammaproteobacteria;Pseudomonadales;Pseudomonadaceae;Pseudomonas;Pseudomonas sp."              
## [58] "Bacteria;Proteobacteria;Gammaproteobacteria;Xanthomonadales;Xanthomonadaceae;Luteimonas;unknown species"               
## [59] "Bacteria;Proteobacteria;Gammaproteobacteria;Xanthomonadales;Xanthomonadaceae;Stenotrophomonas;Stenotrophomonas sp."    
## [60] "Bacteria;Proteobacteria;Gammaproteobacteria;Xanthomonadales;Xanthomonadaceae;Stenotrophomonas;unknown species"         
## [61] "Bacteria;Tenericutes;Mollicutes;Mycoplasmatales;Mycoplasmataceae;Mycoplasma;Mycoplasma bovirhinis"

as well as some of the bast id themselves

unique(names(which(table(df_abundances[ ,2]) > 1)))
##  [1] "AB680687.1.1434" "AB681778.1.1454" "AF224286.1.1534"
##  [4] "AJ491302.1.1509" "AM183097.1.1441" "AY243344.1.1501"
##  [7] "AY725811.1.1474" "FJ672272.1.1408" "FJ674571.1.1375"
## [10] "FJ674662.1.1457" "GQ358834.1.1513" "GQ358846.1.1455"
## [13] "HM325988.1.1341" "HM327870.1.1341" "JX986968.1.1511"
## [16] "KC894531.1.1498"

Microbiote analysis

Preprocessing and normalization

First, the two technical replicates are merged (simple sums as the counts have already been normalized to identical library sizes)

abundances <- df_abundances[ ,grep("A_", colnames(df_abundances))] +
  df_abundances[ ,grep("B_", colnames(df_abundances))]
dim(abundances)
## [1] 406  45

which leads to 45 samples (columns) in which 406 bacteria have been observed (rows).

Condition (“EN” or “LBA”) is extracted from column names:

condition <- rep("LBA", ncol(abundances))
condition[grep("EN", colnames(abundances))] <- "EN"
condition <- factor(condition)
table(condition)
## condition
##  EN LBA 
##  22  23

Also, the farm identifier is extracted from column names:

id_abundances <- as.character(colnames(abundances))
id_abundances <- sapply(id_abundances, function(ac) 
  substr(ac, nchar(ac) - 19, nchar(ac) - 18))
id_abundances <- gsub("A", "0", id_abundances)
id_abundances <- gsub("N", "0", id_abundances)
table(id_abundances)
## id_abundances
## 01 03 04 06 07 09 10 11 15 16 17 18 19 20 21 23 24 25 26 28 29 30 31 
##  2  2  2  2  2  2  2  2  2  2  2  2  2  2  2  2  2  2  2  2  1  2  2

All but one farm (idenfier: 29) have been sampled twice, once for each condition.

Merging duplicates

Duplicated species (based on identical taxonomies) are then merged (counts are summed). To do so, first simplified species names are obtained (last unknown species for every row):

species <- sapply(df_abundances[ ,1], function(aname) 
  unlist(strsplit(aname, ";")))
species <- sapply(species, function(avect) {
  find_unknown <- grep("unknown", avect)
  if (length(find_unknown) > 0) {
    return(avect[-find_unknown])
  } else return(avect)
})
species <- unlist(sapply(species, function(avect) avect[length(avect)]))
species <- unname(species)

Then, counts corresponding to the same simplified name are summed:

abundances <- apply(abundances, 2, function(acol) tapply(acol, species, sum))

and the species names are retrieved:

species <- rownames(abundances)

One species has an unexpected name:

head(species)
## [1] "&"                                   
## [2] "Acetobacteraceae bacterium SAP1007.2"
## [3] "Acholeplasma laidlawii PG-8A"        
## [4] "Achromobacter sp."                   
## [5] "Achromobacter spanius"               
## [6] "Actinoalloteichus cyanogriseus"

that corresponds to the row number 73 in the original data (checked also directly in the file sent by Elias on 10/09).

Exploratory analysis: distribution of one sample

The effect of different normalization is first explored by analyzing the distributions of the counts in the first sample before and after normalization. Distribution before normalization is provided as:

df <- data.frame(abundances)
names(df) <- paste0("sample", 1:ncol(df))
ggplot(df, aes(x = sample1 +1)) + geom_histogram(bins = 50) + scale_y_log10() +
  theme_bw() + xlab("counts (sample 1)") + ggtitle("Sample 1 distribution in raw data")
## Warning: Transformation introduced infinite values in continuous y-axis
## Warning: Removed 41 rows containing missing values (geom_bar).

and with a log-transformation by

ggplot(df, aes(x = sample1 + 1)) + geom_histogram(bins = 50) +
  theme_bw() + xlab("counts (sample 1)") + scale_x_log10() +
  ggtitle("Sample 1 distribution (log scale)")

It is commun in metagenomic datasets to perform TSS (Total Sum Scaling) before further normalization. TSS transformation computes relative abundances: \[ y_{ij} = \frac{n_{ij}}{\sum_{k=1}^p n_{ik}} \] for \(n_{ij}\) the counts of species \(j\) in sample \(i\), \(p\) the number of species and \(n\) the number of individuals.

abundances_TSS <- apply(abundances, 1, function(asample)  asample / sum(asample))
df <- as.data.frame(t(abundances_TSS))
names(df) <- paste0("sample", 1:ncol(df))
dim(df)
## [1] 270  45
ggplot(df, aes(x = sample1 + 1)) + geom_histogram(bins = 50) +
  theme_bw() + xlab("relative abundance (sample 1)") + scale_x_log10() + ggtitle("Sample 1 distribution (TSS)")

The next two histograms are based on the normalized counts with:

  • CLR (Centered Log Ratio) transformation: \[ \tilde{y}_{ij} = \log \frac{y_{ij}}{\sqrt[p]{\prod_{k=1}^p y_{ik}}}. \]
abundances_CLR <- logratio.transfo(abundances_TSS, logratio = "CLR", 
                                   offset = 1)
class(abundances_CLR) <- "matrix"
df <- data.frame(t(abundances_CLR))   
names(df) <- paste0("sample", 1:ncol(df))
dim(df)
## [1] 270  45
ggplot(df, aes(x = sample1)) + geom_histogram(bins = 50) + theme_bw() + 
  xlab("counts (sample 1)") + ggtitle("Sample 1 distribution (CLR)")

  • ILR (Isometric Log Ratio) transformation \[ \tilde{\mathbf{Y}}' = \tilde{\mathbf{Y}} \times \mathbf{V} \] for \(\tilde{\mathbf{Y}}\) the matrix of CLR transformed data and a given matrix \(\mathbf{V}\) with \(p\) rows and \(p-1\) columns such that \(\mathbf{V} \mathbf{V}^\top = \mathbb{I}_{p-1}\) and \(\mathbf{V}^\top \mathbf{V} = \mathbb{I} + a \mathbf{1}\), \(a\) being any positive number and \(\mathbf{1}\) a vector full of 1.
abundances_ILR <- logratio.transfo(abundances_TSS, logratio = "ILR", 
                                   offset = 1)
class(abundances_ILR) <- "matrix"
df <- data.frame(t(abundances_ILR)) 
names(df) <- paste0("sample", 1:ncol(df))
dim(df)
## [1] 269  45
ggplot(df, aes(x = sample1)) + geom_histogram(bins = 50) + theme_bw() + 
  xlab("counts (sample 1)") + ggtitle("Sample 1 distribution (ILR)")

  • CSS transformation, which is an adaptative extension for metagenomic data of the quantile normalization used in microarray expression datasets. It is designed so as to account for technical differences between samples.
abundances_CSS <- newMRexperiment(abundances)
abundances_CSS <- cumNorm(abundances_CSS)
## Default value being used.
df <- data.frame(MRcounts(abundances_CSS))
names(df) <- paste0("sample", 1:ncol(df))
dim(df)
## [1] 270  45
ggplot(df, aes(x = sample1 + 1)) + geom_histogram(bins = 50) + theme_bw() + 
  xlab("counts (sample 1)") + scale_y_log10() + scale_x_log10()
## Warning: Transformation introduced infinite values in continuous y-axis
## Warning: Removed 16 rows containing missing values (geom_bar).

The less asymetric distribution seems to be the one obtained with the CLR transformation and the log-transformed CSS.

Exploratory analysis: distribution of all samples

Distributions of all samples according to the type of transformation and the sample is provided below:

df_log <- log10(abundances + 1)
df_log <- data.frame(df_log)
rownames(df_log) <- NULL
names(df_log) <- paste0("Sample", 1:ncol(df_log))
df_log <- melt(df_log)
df_CLR <- data.frame(t(abundances_CLR))
names(df_CLR) <- paste0("Sample", 1:ncol(df_CLR))
df_CLR <- melt(df_CLR)
df_ILR <- data.frame(t(abundances_ILR))
names(df_ILR) <- paste0("Sample", 1:ncol(df_ILR))
df_ILR <- melt(df_ILR)
df_CSS <- data.frame(log(MRcounts(abundances_CSS)) + 1)
names(df_CSS) <- paste0("Sample", 1:ncol(df_CSS))
df_CSS <- melt(df_CSS)
all_sizes <- c(nrow(df_log), nrow(df_CLR), nrow(df_ILR), nrow(df_CSS))
df <- data.frame(rbind(df_log, df_CLR, df_ILR, df_CSS),
                 "type" = rep(c("log", "CLR", "ILR", "log-CSS"), all_sizes))
ggplot(df, aes(x = variable, y = value)) + geom_boxplot() + theme_bw() +
  facet_wrap(~ type, scales = "free_y") + xlab("samples") +
  theme(axis.text.x = element_blank())
## Warning: Removed 6381 rows containing non-finite values (stat_boxplot).

Exploratory analysis: PCA

Log

A first exploratory analysis is performed with PCA on (merged) raw counts with log transformation:

pca_raw <- pca(log(t(abundances) + 1), ncomp = ncol(abundances), 
               logratio = 'none')
plot(pca_raw)

that shows a good percentage of explained variance for the first axis.

Projection of the individuals on the first two PCs also shows a good separation between the two conditions:

plotIndiv(pca_raw, 
          comp = c(1,2),
          pch = 16, 
          ind.names = FALSE, 
          group = condition, 
          col.per.group = color.mixo(1:2),
          legend = TRUE,
          title="PCA for log-tranformed data")

TSS+CLR

The same analysis is used with TSS normalized counts subsequently transformed by CLR or ILR (which is the expected analysis):

pca_CLR <- pca(abundances_TSS + 1, ncomp = nrow(abundances_TSS),
               logratio = 'CLR')
plot(pca_CLR)

plotIndiv(pca_CLR, 
          comp = c(1,2),
          pch = 16, 
          ind.names = FALSE, 
          group = condition, 
          col.per.group = color.mixo(1:2),
          legend = TRUE,
          title="PCA for CLR-tranformed data")

TSS+ILR

pca_ILR <- pca(abundances_TSS + 1, ncomp = nrow(abundances_TSS) - 1,
               logratio = 'ILR')
plot(pca_ILR)

plotIndiv(pca_ILR, 
          comp = c(1,2),
          pch = 16, 
          ind.names = FALSE, 
          group = condition, 
          col.per.group = color.mixo(1:2),
          legend = TRUE,
          title="PCA for ILR-tranformed data")

CSS

log_CSS <- log(MRcounts(abundances_CSS) + 1)
pca_CSS <- pca(t(log_CSS), ncomp = ncol(log_CSS))
plot(pca_CSS)

plotIndiv(pca_CSS, 
          comp = c(1,2),
          pch = 16, 
          ind.names = FALSE, 
          group = condition, 
          col.per.group = color.mixo(1:2),
          legend = TRUE,
          title="PCA for CSS-tranformed data")

Differences between the two types of samples

A first PLS-DA is computed (with 10-fold CV) to check the efficiency of the method and which type of distance to use in its computation.

with log transformation

set.seed(11)
res_plsda <- plsda(log(t(abundances)+1), condition, ncomp = nlevels(condition))
res_perf <- perf(res_plsda, validation = 'Mfold', folds = 5,
                  progressBar = FALSE, nrepeat = 20)
plot(res_perf, overlay = 'measure', sd = TRUE)

plotIndiv(res_plsda , comp = c(1, 2), ind.names = FALSE, ellipse = TRUE, 
          legend = TRUE, title = 'PLS-DA Comp 1 - 2')

PLS-DA shows a good separation between the two groups and indicates that the Mahalanobis distance provides the lower overall classification error.

Then, sparse PLS-DA is used (with the multilevel approach) to check which number of components to select.

clean_log <- data.frame(log(t(abundances[ ,id_abundances != "29"]) + 1))
names(clean_log) <- species
clean_condition <- factor(condition[id_abundances != "29"])
clean_id <- factor(id_abundances[id_abundances != "29"])

set.seed(33)
res_plsda <- tune.splsda(clean_log, clean_condition, 
                         ncomp = nlevels(clean_condition),
                         multilevel = clean_id,
                         test.keepX = 1:20, validation = 'Mfold', folds = 10, 
                         dist = 'mahalanobis.dist', nrepeat = 10,
                         progressBar = FALSE)
## Splitting the variation for 1 level factor.
plot(res_plsda)

sel_keepX <- res_plsda$choice.keepX[1:2]
sel_keepX
## comp1 comp2 
##     5    14

Finally sparse PLS-DA is performed and the variables explaining the two types of samples are obtained:

res_splsda <- splsda(clean_log, clean_condition, 
                     ncomp = nlevels(clean_condition), multilevel = clean_id,
                     keepX = sel_keepX)
## Splitting the variation for 1 level factor.
plotIndiv(res_splsda, comp = c(1,2), ind.names = FALSE, ellipse = TRUE, 
          legend = TRUE, title = 'sPLS-DA Comp 1 - 2')

head(selectVar(res_splsda, comp = 1)$name)
## [1] "Mycoplasma bovoculi M165/69"    "Corynebacterium camporealensis"
## [3] "Moraxella"                      "Aquamicrobium"                 
## [5] "Peptoclostridium"
plotLoadings(res_splsda, comp = 1, method = 'mean', contrib = 'max',
             size.title = 1)

plotLoadings(res_splsda, comp = 2, method = 'mean', contrib = 'max',
             size.title = 1)

with TSS+CLR transformation

abundances_CLR <- data.frame(abundances_CLR)
names(abundances_CLR) <- species

set.seed(11)
res_plsda <- plsda(abundances_CLR, condition, ncomp = nlevels(condition))
res_perf <- perf(res_plsda, validation = 'Mfold', folds = 5,
                 progressBar = FALSE, nrepeat = 20)
plot(res_perf, overlay = 'measure', sd = TRUE)
plotIndiv(res_plsda , comp = c(1, 2), ind.names = FALSE, ellipse = TRUE, 
          legend = TRUE, title = 'PLS-DA (CLR)')
clean_CLR <- data.frame(abundances_CLR[id_abundances != "29", ])
names(clean_CLR) <- species

set.seed(33)
res_plsda <- tune.splsda(clean_CLR, clean_condition, 
                         ncomp = nlevels(clean_condition),
                         multilevel = clean_id,
                         test.keepX = 1:20, validation = 'Mfold', 
                         folds = 10, dist = 'mahalanobis.dist', nrepeat = 10,
                         progressBar = FALSE)

plot(res_plsda)

sel_keepX <- res_plsda$choice.keepX[1:2]
sel_keepX
res_splsda <- splsda(clean_CLR, clean_condition, 
                     ncomp = nlevels(clean_condition), multilevel = clean_id,
                     keepX = sel_keepX)

plotIndiv(res_splsda, comp = c(1,2), ind.names = FALSE, ellipse = TRUE, 
          legend = TRUE, title = 'sPLS-DA (CLR)')
plotLoadings(res_splsda, comp = 1, method = 'mean', contrib = 'max')
plotLoadings(res_splsda, comp = 2, method = 'mean', contrib = 'max')

with CSS transformation

abundances_CSS <- data.frame(t(log_CSS))
names(abundances_CSS) <- species

set.seed(11)
res_plsda <- plsda(abundances_CSS, condition, ncomp = nlevels(condition))
res_perf <- perf(res_plsda, validation = 'Mfold', folds = 5,
                 progressBar = FALSE, nrepeat = 20)
plot(res_perf, overlay = 'measure', sd = TRUE)

plotIndiv(res_plsda , comp = c(1, 2), ind.names = FALSE, ellipse = TRUE, 
          legend = TRUE, title = 'PLS-DA (CSS)')

clean_CSS <- data.frame(abundances_CSS[id_abundances != "29", ])
names(clean_CSS) <- species

set.seed(33)
res_plsda <- tune.splsda(clean_CSS, clean_condition, 
                         ncomp = nlevels(clean_condition),
                         multilevel = clean_id,
                         test.keepX = 1:20, validation = 'Mfold', 
                         folds = 10, dist = 'mahalanobis.dist', nrepeat = 10,
                         progressBar = FALSE)
## Splitting the variation for 1 level factor.
plot(res_plsda)

sel_keepX <- res_plsda$choice.keepX[1:2]
sel_keepX
## comp1 comp2 
##     5    14
res_splsda <- splsda(clean_CSS, clean_condition, 
                     ncomp = nlevels(clean_condition), multilevel = clean_id,
                     keepX = sel_keepX)
## Splitting the variation for 1 level factor.
plotIndiv(res_splsda, comp = c(1,2), ind.names = FALSE, ellipse = TRUE, 
          legend = TRUE, title = 'sPLS-DA (CSS)')

plotLoadings(res_splsda, comp = 1, method = 'mean', contrib = 'max')

plotLoadings(res_splsda, comp = 2, method = 'mean', contrib = 'max')

with log transformation (but not paired)

set.seed(33)
res_plsda <- tune.splsda(log(t(abundances)+1), condition, 
                         ncomp = nlevels(condition), 
                         test.keepX = 1:20, validation = 'Mfold', 
                         folds = 10, dist = 'mahalanobis.dist', nrepeat = 10,
                         progressBar = FALSE)

plot(res_plsda)

sel_keepX <- res_plsda$choice.keepX[1:2]
sel_keepX
## comp1 comp2 
##    20    15
res_splsda <- splsda(log(t(abundances)+1), condition, 
                     ncomp = nlevels(condition), keepX = sel_keepX)

plotIndiv(res_splsda, comp = c(1,2), ind.names = FALSE, ellipse = TRUE, 
          legend = TRUE, title = 'sPLS-DA (log)')

plotLoadings(res_splsda, comp = 1, method = 'mean', contrib = 'max',
             size.title = 1)

plotLoadings(res_splsda, comp = 2, method = 'mean', contrib = 'max',
             size.title = 1)

with TSS+CLR transformation (not paired)

set.seed(11)
res_plsda <- plsda(abundances_CLR, condition, ncomp = nlevels(condition))
res_perf <- perf(res_plsda, validation = 'Mfold', folds = 5, 
                 progressBar = FALSE, nrepeat = 20)
plot(res_perf, overlay = 'measure', sd = TRUE)
plotIndiv(res_plsda , comp = c(1, 2), ind.names = FALSE, ellipse = TRUE, 
          legend = TRUE, title = 'PLS-DA (CLR)')
set.seed(33)
res_plsda <- tune.splsda(abundances_CLR, condition, ncomp = nlevels(condition),
                         test.keepX = 1:20, validation = 'Mfold', 
                         folds = 10, dist = 'mahalanobis.dist', nrepeat = 10,
                         progressBar = F)

plot(res_plsda)

sel_keepX <- res_plsda$choice.keepX[1:2]
sel_keepX
res_splsda <- splsda(abundances_CLR, condition, ncomp = nlevels(condition),
                     keepX = sel_keepX)

plotIndiv(res_splsda, comp = c(1,2), ind.names = FALSE, ellipse = TRUE, 
          legend = TRUE, title = 'sPLS-DA (CLR)')
plotLoadings(res_splsda, comp = 1, method = 'mean', contrib = 'max',
             size.title = 1)
plotLoadings(res_splsda, comp = 2, method = 'mean', contrib = 'max',
             size.title = 1)

Random Forest

In this section will be performed Random Forest algorithm for classification, with a virus as response vector and all bacteria as explanatory variables in the model.

library(randomForest,quietly=T)

Data preparation:

df_pathogenes<-read.delim("../data/pathogenes.csv",sep = ",")
pathogenes<-df_pathogenes[,-c(1,9)]
pathogenes<-as.data.frame(ifelse(pathogenes=='p',1,0))
pathogenes<-as.data.frame(apply(t(pathogenes),1,as.factor))
id_pathogenes<-df_pathogenes[,1]
id_pathogenes <- gsub("-","0",id_pathogenes)
id_pathogenes <- gsub(" ","",id_pathogenes)
id_pathogenes <- sapply(as.character(id_pathogenes),FUN=function(x){substr(x,nchar(x)-1,nchar(x))})
id_pathogenes <- id_pathogenes[id_pathogenes!="29"]
id_pathogenes <- paste0(id_pathogenes,"_",rev(condition[id_abundances!="29"]))

The list of calf names is not in the same order in the pathogenes file as in the abundances file. The abundances file order will be kept:

id_abundances_bis <- paste0(id_abundances[-grep("29",id_abundances)],"_",
                            condition[-grep("29",id_abundances)])
Y<-pathogenes[match(id_abundances_bis,id_pathogenes),] 
rownames(Y)<-id_abundances_bis

Y contains the seven response vectors.

cat(" 0  1\n\n")
##  0  1
for (virus in colnames(Y)){
  cat(virus,"\n",table(Y[,virus]),fill=T)
}
## Ct.RSV 
##  37 7
## Ct.PI.3 
##  41 3
## Ct.Coronavirus 
##  22 22
## Ct.P.multocida 
##  7 37
## Ct.M.haemolytica 
##  33 11
## Ct.M.bovis 
##  40 4
## Ct.H.somni 
##  27 17

The bacteria names are too long for the plots, they will be renamed:

X <- clean_log
X <- as.data.frame(sapply(X,as.integer))
colnames(X)<- make.names(names(X))

Random Forest on both conditions

Random Forest on RSV:

fit <- randomForest(Y$Ct.RSV ~ ., data = X,classwt=table(Y$Ct.RSV),ntree=3000,mtry=10)
fit$confusion
##    0 1 class.error
## 0 37 0           0
## 1  7 0           1

Nothing in class 1…

varImpPlot(fit)
round(fit$importance[order(fit$importance[, 1], decreasing = TRUE), ][1:5],3)
plot(fit$err.rate[, 1], type = "l", xlab = "number of trees", ylab = "error OOB")
cat(fit$predicted)

Random Forest on PI.3:

fit <- randomForest(Y$Ct.PI.3 ~ ., data = X,classwt=table(Y$Ct.PI.3),ntree=1000,mtry=10)
fit$confusion
##    0 1 class.error
## 0 41 0           0
## 1  3 0           1

Nothing in class 1…

varImpPlot(fit)
round(fit$importance[order(fit$importance[, 1], decreasing = TRUE), ][1:5],3)
plot(fit$err.rate[, 1], type = "l", xlab = "number of trees", ylab = "error OOB")

Random Forest on Coronavirus:

fit <- randomForest(Y$Ct.Coronavirus ~ .,data = X,classwt=table(Y$Ct.Coronavirus)
                    , ntree=2000,mtry=16)
fit$confusion
##    0  1 class.error
## 0 10 12   0.5454545
## 1 11 11   0.5000000

High error

varImpPlot(fit)
fit$importance[order(fit$importance[, 1], decreasing = TRUE), ][1:5]
plot(fit$err.rate[, 1], type = "l", xlab = "number of trees", ylab = "error OOB")

Random Forest on P.multocida:

fit <- randomForest(Y$Ct.P.multocida ~ ., data = X,classwt=table(Y$Ct.P.multocida),ntree=1000,mtry=16)
fit$confusion
##   0  1 class.error
## 0 0  7           1
## 1 0 37           0

Nothing in class 0

varImpPlot(fit)
fit$importance[order(fit$importance[, 1], decreasing = TRUE), ][1:5]
plot(fit$err.rate[, 1], type = "l", xlab = "nombre d'arbres", ylab = "erreur OOB")

Random Forest on M.haemolytica:

fit <- randomForest(Y$Ct.M.haemolytica ~ ., data = X,classwt=table(Y$Ct.M.haemolytica),ntree=1000,mtry=40)
fit$confusion
##    0 1 class.error
## 0 31 2  0.06060606
## 1 11 0  1.00000000
varImpPlot(fit)
fit$importance[order(fit$importance[, 1], decreasing = TRUE), ][1:5]
plot(fit$err.rate[, 1], type = "l", xlab = "number of trees", ylab = "error OOB")

Random Forest on M.bovis:

fit <- randomForest(Y$Ct.M.bovis ~ ., data = X,classwt=table(Y$Ct.M.bovis) , ntree=500,mtry=40)
fit$confusion
##    0 1 class.error
## 0 40 0           0
## 1  4 0           1
varImpPlot(fit)
fit$importance[order(fit$importance[, 1], decreasing = TRUE), ][1:5]
plot(fit$err.rate[, 1], type = "l", xlab = "number of trees", ylab = "error OOB")

Random Forest on H.somni:

fit <- randomForest(Y$Ct.H.somni ~ ., data = X,classwt=table(Y$Ct.H.somni),ntree=1000,mtry=16)
fit$confusion
##    0 1 class.error
## 0 24 3   0.1111111
## 1 17 0   1.0000000
varImpPlot(fit)
fit$importance[order(fit$importance[, 1], decreasing = TRUE), ][1:5]
plot(fit$err.rate[, 1], type = "l", xlab = "number of trees", ylab = "error OOB")

Random Forest on EN condition

Study only on EN condition samples

X_EN<-X[grep("EN",colnames(abundances)),]
X_LBA<-X[grep("LBA",rownames(Y)),]
Y_EN<-Y[grep("EN",colnames(abundances)),]
Y_LBA<-Y[grep("LBA",rownames(Y)),]

clean_log_EN <- clean_log[grep("EN",rownames(clean_log)),]
clean_id_EN <- clean_id[grep("EN",names(clean_id))]
clean_log_LBA <- clean_log[grep("LBA",rownames(clean_log)),]
clean_id_LBA <- clean_id[grep("LBA",names(clean_id))]

Random Forest on RSV - EN condition:

fit <- randomForest(Y_EN$Ct.RSV ~ ., data = X_EN,classwt=table(Y_EN$Ct.H.somni), ntree=5000,mtry=40)
fit$confusion
##    0 1 class.error
## 0 20 0           0
## 1  2 0           1

Random Forest on PI.3 - EN condition:

fit <- randomForest(Y_EN$Ct.PI.3 ~ ., data = X_EN,classwt=table(Y_EN$Ct.H.somni), ntree=5000,mtry=40)
fit$confusion
##    0 1 class.error
## 0 20 0           0
## 1  2 0           1

Random Forest on Coronavirus - EN condition:

fit <- randomForest(Y_EN$Ct.Coronavirus ~ ., data = X_EN,classwt=table(Y_EN$Ct.H.somni), ntree=5000,mtry=40)
fit$confusion
##   0  1 class.error
## 0 0  9   1.0000000
## 1 2 11   0.1538462

Random Forest on M.haemolytica - EN condition:

fit <- randomForest(Y_EN$Ct.M.haemolytica~ ., data = X_EN,classwt=table(Y_EN$Ct.H.somni), ntree=5000,mtry=40)
fit$confusion
##    0 1 class.error
## 0 14 2       0.125
## 1  6 0       1.000

Random Forest on M.bovis - EN condition:

fit <- randomForest(Y_EN$Ct.M.bovis~ ., data = X_EN,classwt=table(Y_EN$Ct.H.somni), ntree=5000,mtry=40)
fit$confusion
##    0 1 class.error
## 0 21 0           0
## 1  0 0         NaN

Random Forest on P.multocida - EN condition:

fit <- randomForest(Y_EN$Ct.P.multocida ~ ., data = X_EN,classwt=table(Y_EN$Ct.H.somni), ntree=5000,mtry=40)
fit$confusion
##   0  1 class.error
## 0 0  2           1
## 1 0 20           0

Random Forest on H.somni - EN condition:

fit <- randomForest(Y_EN$Ct.H.somni ~ ., data = X_EN,classwt=table(Y_EN$Ct.H.somni), ntree=5000,mtry=40)
fit$confusion
##    0 1 class.error
## 0 12 2   0.1428571
## 1  8 0   1.0000000

PLS-DA on viruses

PLS-DA on RSV

set.seed(11)
res_plsda <- plsda(X=clean_log,Y=Y$Ct.RSV, ncomp = nlevels(clean_condition))
res_perf <- perf(res_plsda, validation = 'Mfold', folds = 5,
                 progressBar = FALSE, nrepeat = 20)
plot(res_perf, overlay = 'measure', sd = TRUE)

plotIndiv(res_plsda , comp = c(1, 2), ind.names = FALSE, ellipse = TRUE, 
          legend = TRUE, title = 'PLS-DA (RSV)')

set.seed(33)
res_plsda <- tune.splsda(X=clean_log,Y=Y$Ct.RSV, 
                         ncomp = nlevels(clean_condition),
                         multilevel = clean_id,
                         test.keepX = 1:20, validation = 'Mfold', 
                         folds = 10, dist = 'mahalanobis.dist', nrepeat = 10,
                         progressBar = F)
## Splitting the variation for 1 level factor.
plot(res_plsda)

sel_keepX <- res_plsda$choice.keepX[1:2]
sel_keepX
## comp1 comp2 
##    13    14
res_splsda <- splsda(X=clean_log,Y=Y$Ct.RSV, 
                     ncomp = nlevels(clean_condition),
                     multilevel = clean_id,
                     keepX = sel_keepX)
## Splitting the variation for 1 level factor.
plotIndiv(res_splsda, comp = c(1,2), ind.names = FALSE, ellipse = TRUE, 
          legend = TRUE, title = 'sPLS-DA (RSV)')

Si on limite le nombre de variables, on n’arrive pas a discrimner de maniere lineaire la presence ou l’absence du Coronavirus.

head(selectVar(res_splsda, comp = 1)$name)
## [1] "Streptococcus gallolyticus"                   
## [2] "Staphylococcus cohnii"                        
## [3] "Prevotella heparinolytica"                    
## [4] "Ureaplasma"                                   
## [5] "cilia-associated respiratory bacterium 246-57"
## [6] "Saccharibacteria"
plotLoadings(res_splsda, comp = 1, method = 'mean', contrib = 'max',
             size.title = 1)

plotLoadings(res_splsda, comp = 2, method = 'mean', contrib = 'max',
             size.title = 1)

PLS-DA on PI.3

set.seed(11)
res_plsda <- plsda(X=clean_log,Y=Y$Ct.PI.3, ncomp = nlevels(clean_condition))
res_perf <- perf(res_plsda, validation = 'Mfold', folds = 5,
                 progressBar = FALSE, nrepeat = 20)
## Warning in MCVfold.splsda(X, Y, multilevel = multilevel, validation = validation, : At least one class is not represented in one fold, which may unbalance the error rate.
##   Consider a number of folds lower than the minimum in table(Y): 3

## Warning in MCVfold.splsda(X, Y, multilevel = multilevel, validation = validation, : At least one class is not represented in one fold, which may unbalance the error rate.
##   Consider a number of folds lower than the minimum in table(Y): 3
plot(res_perf, overlay = 'measure', sd = TRUE)

plotIndiv(res_plsda , comp = c(1, 2), ind.names = FALSE, ellipse = TRUE, 
          legend = TRUE, title = 'PLS-DA (PI.3)')

set.seed(33)
res_plsda <- tune.splsda(X=clean_log,Y=Y$Ct.PI.3, 
                         ncomp = nlevels(clean_condition),
                         multilevel = clean_id,
                         test.keepX = 1:20, validation = 'Mfold', 
                         folds = 10, dist = 'mahalanobis.dist', nrepeat = 10,
                         progressBar = F)
## Splitting the variation for 1 level factor.
plot(res_plsda)

sel_keepX <- res_plsda$choice.keepX[1:2]
sel_keepX
## comp1 comp2 
##    11    16
res_splsda <- splsda(X=clean_log,Y=Y$Ct.PI.3, 
                     ncomp = nlevels(clean_condition),
                     multilevel = clean_id,
                     keepX = sel_keepX)
## Splitting the variation for 1 level factor.
plotIndiv(res_splsda, comp = c(1,2), ind.names = FALSE, ellipse = TRUE, 
          legend = TRUE, title = 'sPLS-DA (PI.3)')

head(selectVar(res_splsda, comp = 1)$name)
## [1] "Enterococcus hirae ATCC 9790"     "&"                               
## [3] "Escherichia coli"                 "Ulvibacter"                      
## [5] "Pseudomonas aeruginosa"           "Lactobacillus ruminis ATCC 27782"
plotLoadings(res_splsda, comp = 1, method = 'mean', contrib = 'max',
             size.title = 1)

plotLoadings(res_splsda, comp = 2, method = 'mean', contrib = 'max',
             size.title = 1)

PLS-DA on Coronavirus

set.seed(11)
res_plsda <- plsda(X=clean_log,Y=Y$Ct.Coronavirus, ncomp = nlevels(clean_condition))
res_perf <- perf(res_plsda, validation = 'Mfold', folds = 5,
                 progressBar = FALSE, nrepeat = 20)
plot(res_perf, overlay = 'measure', sd = TRUE)

plotIndiv(res_plsda , comp = c(1, 2), ind.names = FALSE, ellipse = TRUE, 
          legend = TRUE, title = 'PLS-DA (Coronavirus)')

set.seed(33)
res_plsda <- tune.splsda(X=clean_log,Y=Y$Ct.Coronavirus, 
                         ncomp = nlevels(clean_condition),
                         multilevel = clean_id,
                         test.keepX = 1:20, validation = 'Mfold', 
                         folds = 10, dist = 'mahalanobis.dist', nrepeat = 10,
                         progressBar = F)
## Splitting the variation for 1 level factor.
plot(res_plsda)

sel_keepX <- res_plsda$choice.keepX[1:2]
sel_keepX
## comp1 comp2 
##    13    14
res_splsda <- splsda(X=clean_log,Y=Y$Ct.Coronavirus, 
                     ncomp = nlevels(clean_condition),
                     multilevel = clean_id,
                     keepX = sel_keepX)
## Splitting the variation for 1 level factor.
plotIndiv(res_splsda, comp = c(1,2), ind.names = FALSE, ellipse = TRUE, 
          legend = TRUE, title = 'sPLS-DA (Coronavirus)')

Si on limite le nombre de variables, on n’arrive pas a discrimner de maniere lineaire la presence ou l’absence du Coronavirus.

head(selectVar(res_splsda, comp = 1)$name)
## [1] "Bacteroides"                "Parvimonas"                
## [3] "Family XIII"                "Methylobacterium sp."      
## [5] "Alloprevotella"             "Streptomyces tempisquensis"
plotLoadings(res_splsda, comp = 1, method = 'mean', contrib = 'max',
             size.title = 1)

plotLoadings(res_splsda, comp = 2, method = 'mean', contrib = 'max',
             size.title = 1)

PLS-DA on P.multocida

set.seed(11)
res_plsda <- plsda(X=clean_log,Y=Y$Ct.P.multocida, ncomp = nlevels(clean_condition))
res_perf <- perf(res_plsda, validation = 'Mfold', folds = 5,
                 progressBar = FALSE, nrepeat = 20)
plot(res_perf, overlay = 'measure', sd = TRUE)

plotIndiv(res_plsda , comp = c(1, 2), ind.names = FALSE, ellipse = TRUE, 
          legend = TRUE, title = 'PLS-DA (P.multocida)')

set.seed(33)
res_plsda <- tune.splsda(X=clean_log,Y=Y$Ct.P.multocida, 
                         ncomp = nlevels(clean_condition),
                         multilevel = clean_id,
                         test.keepX = 1:20, validation = 'Mfold', 
                         folds = 10, dist = 'mahalanobis.dist', nrepeat = 10,
                         progressBar = F)
## Splitting the variation for 1 level factor.
plot(res_plsda)

sel_keepX <- res_plsda$choice.keepX[1:2]
sel_keepX
## comp1 comp2 
##     1    16
res_splsda <- splsda(X=clean_log,Y=Y$Ct.P.multocida, 
                     ncomp = nlevels(clean_condition),
                     multilevel = clean_id,
                     keepX = sel_keepX)
## Splitting the variation for 1 level factor.
plotIndiv(res_splsda, comp = c(1,2), ind.names = FALSE, ellipse = TRUE, 
          legend = TRUE, title = 'sPLS-DA (P.multocida)')

head(selectVar(res_splsda, comp = 1)$name)
## [1] "Streptomyces sp."
plotLoadings(res_splsda, comp = 1, method = 'mean', contrib = 'max',
             size.title = 1)

plotLoadings(res_splsda, comp = 2, method = 'mean', contrib = 'max',
             size.title = 1)

PLS-DA on M.haemolytica

set.seed(11)
res_plsda <- plsda(X=clean_log,Y=Y$Ct.M.haemolytica, ncomp = nlevels(clean_condition))
res_perf <- perf(res_plsda, validation = 'Mfold', folds = 5,
                 progressBar = FALSE, nrepeat = 20)
plot(res_perf, overlay = 'measure', sd = TRUE)

plotIndiv(res_plsda , comp = c(1, 2), ind.names = FALSE, ellipse = TRUE, 
          legend = TRUE, title = 'PLS-DA (M.haemolytica)')

set.seed(33)
res_plsda <- tune.splsda(X=clean_log,Y=Y$Ct.M.haemolytica, 
                         ncomp = nlevels(clean_condition),
                         multilevel = clean_id,
                         test.keepX = 1:20, validation = 'Mfold', 
                         folds = 10, dist = 'mahalanobis.dist', nrepeat = 10,
                         progressBar = F)
## Splitting the variation for 1 level factor.
plot(res_plsda)

sel_keepX <- res_plsda$choice.keepX[1:2]
sel_keepX
## comp1 comp2 
##    18    13
res_splsda <- splsda(X=clean_log,Y=Y$Ct.M.haemolytica, 
                     ncomp = nlevels(clean_condition),
                     multilevel = clean_id,
                     keepX = sel_keepX)
## Splitting the variation for 1 level factor.
plotIndiv(res_splsda, comp = c(1,2), ind.names = FALSE, ellipse = TRUE, 
          legend = TRUE, title = 'sPLS-DA (M.haemolytica)')

head(selectVar(res_splsda, comp = 1)$name)
## [1] "Nocardiopsis"               "Nocardiopsis sp."          
## [3] "Streptomyces tempisquensis" "Leptospira broomii"        
## [5] "Saccharopolyspora gregorii" "Saccharomonospora sp."
plotLoadings(res_splsda, comp = 1, method = 'mean', contrib = 'max',
             size.title = 1)

plotLoadings(res_splsda, comp = 2, method = 'mean', contrib = 'max',
             size.title = 1)

PLS-DA on M.bovis

set.seed(11)
res_plsda <- plsda(X=clean_log,Y=Y$Ct.M.bovis, ncomp = nlevels(clean_condition))
res_perf <- perf(res_plsda, validation = 'Mfold', folds = 5,
                 progressBar = FALSE, nrepeat = 20)
## Warning in MCVfold.splsda(X, Y, multilevel = multilevel, validation = validation, : At least one class is not represented in one fold, which may unbalance the error rate.
##   Consider a number of folds lower than the minimum in table(Y): 4

## Warning in MCVfold.splsda(X, Y, multilevel = multilevel, validation = validation, : At least one class is not represented in one fold, which may unbalance the error rate.
##   Consider a number of folds lower than the minimum in table(Y): 4
plot(res_perf, overlay = 'measure', sd = TRUE)

plotIndiv(res_plsda , comp = c(1, 2), ind.names = FALSE, ellipse = TRUE, 
          legend = TRUE, title = 'PLS-DA (M.bovis)')

set.seed(33)
res_plsda <- tune.splsda(X=clean_log,Y=Y$Ct.M.bovis, 
                         ncomp = nlevels(clean_condition),
                         multilevel = clean_id,
                         test.keepX = 1:20, validation = 'Mfold', 
                         folds = 10, dist = 'mahalanobis.dist', nrepeat = 10,
                         progressBar = F)
## Splitting the variation for 1 level factor.
plot(res_plsda)

sel_keepX <- res_plsda$choice.keepX[1:2]
sel_keepX
## comp1 comp2 
##    12     7
res_splsda <- splsda(X=clean_log,Y=Y$Ct.M.bovis, 
                     ncomp = nlevels(clean_condition),
                     multilevel = clean_id,
                     keepX = sel_keepX)
## Splitting the variation for 1 level factor.
plotIndiv(res_splsda, comp = c(1,2), ind.names = FALSE, ellipse = TRUE, 
          legend = TRUE, title = 'sPLS-DA (M.bovis)')

head(selectVar(res_splsda, comp = 1)$name)
## [1] "Helcococcus ovis"           "Arcobacter cryaerophilus"  
## [3] "Thermoactinomyces vulgaris" "Trueperella pyogenes"      
## [5] "Gracilibacteria"            "Saccharopolyspora gregorii"
plotLoadings(res_splsda, comp = 1, method = 'mean', contrib = 'max',
             size.title = 1)

plotLoadings(res_splsda, comp = 2, method = 'mean', contrib = 'max',
             size.title = 1)

PLS-DA on H.somni

set.seed(11)
res_plsda <- plsda(X=clean_log,Y=Y$Ct.H.somni, ncomp = nlevels(clean_condition))
res_perf <- perf(res_plsda, validation = 'Mfold', folds = 5,
                 progressBar = FALSE, nrepeat = 20)
plot(res_perf, overlay = 'measure', sd = TRUE)

plotIndiv(res_plsda , comp = c(1, 2), ind.names = FALSE, ellipse = TRUE, 
          legend = TRUE, title = 'PLS-DA (H.somni)')

set.seed(33)
res_plsda <- tune.splsda(X=clean_log,Y=Y$Ct.H.somni, 
                         ncomp = nlevels(clean_condition),
                         multilevel = clean_id,
                         test.keepX = 1:20, validation = 'Mfold', 
                         folds = 10, dist = 'mahalanobis.dist', nrepeat = 10,
                         progressBar = F)
## Splitting the variation for 1 level factor.
plot(res_plsda)

sel_keepX <- res_plsda$choice.keepX[1:2]
sel_keepX
## comp1 comp2 
##     2     5
res_splsda <- splsda(X=clean_log,Y=Y$Ct.H.somni, 
                     ncomp = nlevels(clean_condition),
                     multilevel = clean_id,
                     keepX = sel_keepX)
## Splitting the variation for 1 level factor.
plotIndiv(res_splsda, comp = c(1,2), ind.names = FALSE, ellipse = TRUE, 
          legend = TRUE, title = 'sPLS-DA (H.somni)')

head(selectVar(res_splsda, comp = 1)$name)
## [1] "Saccharopolyspora gloriosae"      "Lactobacillus ruminis ATCC 27782"
plotLoadings(res_splsda, comp = 1, method = 'mean', contrib = 'max',
             size.title = 1)

plotLoadings(res_splsda, comp = 2, method = 'mean', contrib = 'max',
             size.title = 1)

sPLS with regression mode - EN condition

pls <- pls(as.matrix(sapply(X_EN,as.numeric)), as.matrix(sapply(Y_EN,as.numeric)), ncomp = 8, mode = "regression")
## Warning in cor(A[[k]], variates.A[[k]]): l'écart type est nulle
spls <- spls(as.matrix(sapply(X_EN,as.numeric)), as.matrix(sapply(Y_EN,as.numeric)), ncomp =8, 
             keepX= c(15,15), keepY=c(7,7), mode = "regression")
## Warning in cor(A[[k]], variates.A[[k]]): l'écart type est nulle
tune.pls <- perf(pls, validation = "Mfold", folds = 20, 
                 progressBar = FALSE, nrepeat = 10)
## Warning in cor(A[[k]], variates.A[[k]]): l'écart type est nulle
tune.spls <- perf(spls, validation = "Mfold", folds = 20,
                  progressBar = FALSE, nrepeat = 10)
## Warning in cor(A[[k]], variates.A[[k]]): l'écart type est nulle
tune.pls$Q2.total
##           Q2.total
## 1 comp -0.09134376
## 2 comp -0.06484187
## 3 comp -0.08695912
## 4 comp  0.05612558
## 5 comp  0.05659297
## 6 comp -0.01717743
## 7 comp  0.13082009
## 8 comp -0.01482035
plot(tune.pls$Q2.total)
abline(h=0.0975)

Sample Plots

plotIndiv(spls, comp = 1:2, rep.space= 'Y-variate', ind.names = clean_id[grep("EN",names(clean_id))],
legend = TRUE, title = 'sPLS comp 1 - 2, Y-space')

plotIndiv(spls, comp = 1:2, rep.space= 'X-variate', ind.names = clean_id[grep("EN",names(clean_id))],
legend = TRUE, title = 'sPLS comp 1 - 2, X-space')

plotIndiv(spls, comp = 1:2, rep.space= 'XY-variate', ind.names = clean_id[grep("EN",names(clean_id))],
legend = TRUE, title = 'sPLS comp 1 - 2, XY-space')

Individual plots can be displayed on three different subspaces spanned either by the X variable, the Y variable or the mean subspace in which coordinates are averaged from the first two subspaces (XY).

Variable Plot

plotVar(spls, comp =1:2, var.names = list(X.label = species, 
        Y.label = TRUE), cex = c(4, 5))

sPLS with regression mode - LBA condition

pls <- pls(as.matrix(sapply(X_LBA,as.numeric)), as.matrix(sapply(Y_LBA,as.numeric)), ncomp = 8, mode = "regression")
## Warning in cor(A[[k]], variates.A[[k]]): l'écart type est nulle
spls <- spls(as.matrix(sapply(X_LBA,as.numeric)), as.matrix(sapply(Y_LBA,as.numeric)), ncomp =8, 
             keepX= c(15,15), keepY=c(7,7), mode = "regression")
## Warning in cor(A[[k]], variates.A[[k]]): l'écart type est nulle
tune.pls <- perf(pls, validation = "Mfold", folds = 20, 
                 progressBar = FALSE, nrepeat = 10)
## Warning in cor(A[[k]], variates.A[[k]]): l'écart type est nulle
tune.spls <- perf(spls, validation = "Mfold", folds = 20,
                  progressBar = FALSE, nrepeat = 10)
## Warning in cor(A[[k]], variates.A[[k]]): l'écart type est nulle
## Warning: The SGCCA algorithm did not converge
tune.pls$Q2.total
##           Q2.total
## 1 comp -0.02386096
## 2 comp -0.01564945
## 3 comp -0.00623711
## 4 comp -0.03482518
## 5 comp  0.01454610
## 6 comp  0.07626493
## 7 comp  0.03007753
## 8 comp  0.02123013
plot(tune.pls$Q2.total)
abline(h=0.0975)

Sample Plots

plotIndiv(spls, comp = 1:2, rep.space= 'Y-variate', ind.names = clean_id_LBA,
legend = TRUE, title = 'sPLS comp 1 - 2, Y-space')

plotIndiv(spls, comp = 1:2, rep.space= 'X-variate', ind.names = clean_id_LBA,
legend = TRUE, title = 'sPLS comp 1 - 2, X-space')

plotIndiv(spls, comp = 1:2, rep.space= 'XY-variate', ind.names = clean_id_LBA,
legend = TRUE, title = 'sPLS comp 1 - 2, XY-space')

Individual plots can be displayed on three different subspaces spanned either by the X variable, the Y variable or the mean subspace in which coordinates are averaged from the first two subspaces (XY).

Variable Plot

plotVar(spls, comp =1:2, var.names = list(X.label = species, 
        Y.label = TRUE), cex = c(4, 5))

#cim(spls, comp = 1:2, xlab = "virus", ylab = "genes")

Work on one condition

Regroupment of RSV, PI.3, Coronavirus

group_virus<-sapply(Y,as.numeric)
group_virus<-group_virus-1
group_virus<-group_virus[,1]+group_virus[,2]+group_virus[,3]
group_virus[group_virus>0]<-1
group_virus <- as.factor(group_virus)

group_virus_EN <- group_virus[grep("EN",rownames(Y))]
group_virus_LBA <- group_virus[grep("LBA",rownames(Y))]

PLS-DA on group_virus - EN condition

set.seed(11)
res_plsda <- plsda(X=clean_log_EN,Y=group_virus_EN, ncomp = nlevels(group_virus_EN))
## Warning in cor(A[[k]], variates.A[[k]]): l'écart type est nulle
res_perf <- perf(res_plsda, validation = 'Mfold', folds = 5,
                 progressBar = FALSE, nrepeat = 20)
plot(res_perf, overlay = 'measure', sd = TRUE)

plotIndiv(res_plsda , comp = c(1, 2), ind.names = FALSE, ellipse = TRUE, 
          legend = TRUE, title = 'PLS-DA (RSV,PI.3 et Coronavirus) - EN')

set.seed(33)
res_plsda <- tune.splsda(X=clean_log_EN,Y=group_virus_EN, 
                         ncomp = nlevels(group_virus_EN),
                         test.keepX = 1:20, validation = 'Mfold', 
                         folds = 10, dist = 'mahalanobis.dist', nrepeat = 10,
                         progressBar = F)
## Warning in MCVfold.splsda(X, Y, multilevel = multilevel, validation = validation, : At least one class is not represented in one fold, which may unbalance the error rate.
##   Consider a number of folds lower than the minimum in table(Y): 8

## Warning in MCVfold.splsda(X, Y, multilevel = multilevel, validation = validation, : At least one class is not represented in one fold, which may unbalance the error rate.
##   Consider a number of folds lower than the minimum in table(Y): 8
plot(res_plsda)

sel_keepX <- res_plsda$choice.keepX[1:2]
sel_keepX
## comp1 comp2 
##     1    19
res_splsda <- splsda(X=clean_log_EN,Y=group_virus_EN, 
                     ncomp = nlevels(group_virus_EN),
                     keepX = c(5,19))
## Warning in cor(A[[k]], variates.A[[k]]): l'écart type est nulle
plotIndiv(res_splsda, comp = c(1,2), ind.names = FALSE, ellipse = TRUE, 
          legend = TRUE, title = 'sPLS-DA (RSV,PI.3 et Coronavirus) - EN')

head(selectVar(res_splsda, comp = 1)$name)
## [1] "Pasteurella multocida"      "Staphylococcus chromogenes"
## [3] "Delftia sp."                "Mesorhizobium"             
## [5] "Weeksella"
plotLoadings(res_splsda, comp = 1, method = 'mean', contrib = 'max',
             size.title = 1)

plotLoadings(res_splsda, comp = 2, method = 'mean', contrib = 'max',
             size.title = 1)

PLS-DA on group_virus - LBA condition

set.seed(11)
res_plsda <- plsda(X=clean_log_LBA,Y=group_virus_LBA, ncomp = nlevels(group_virus_LBA))
## Warning in cor(A[[k]], variates.A[[k]]): l'écart type est nulle
res_perf <- perf(res_plsda, validation = 'Mfold', folds = 5,
                 progressBar = FALSE, nrepeat = 20)
plot(res_perf, overlay = 'measure', sd = TRUE)

plotIndiv(res_plsda , comp = c(1, 2), ind.names = FALSE, ellipse = TRUE, 
          legend = TRUE, title = 'PLS-DA (RSV,PI.3 et Coronavirus) - LBA')

set.seed(33)
res_plsda <- tune.splsda(X=clean_log_LBA,Y=group_virus_LBA, 
                         ncomp = nlevels(group_virus_LBA),
                         test.keepX = 1:20, validation = 'Mfold', 
                         folds = 10, dist = 'mahalanobis.dist', nrepeat = 10,
                         progressBar = F)
## Warning in MCVfold.splsda(X, Y, multilevel = multilevel, validation = validation, : At least one class is not represented in one fold, which may unbalance the error rate.
##   Consider a number of folds lower than the minimum in table(Y): 9

## Warning in MCVfold.splsda(X, Y, multilevel = multilevel, validation = validation, : At least one class is not represented in one fold, which may unbalance the error rate.
##   Consider a number of folds lower than the minimum in table(Y): 9
## Warning: The SGCCA algorithm did not converge

## Warning: The SGCCA algorithm did not converge

## Warning: The SGCCA algorithm did not converge
plot(res_plsda)

sel_keepX <- res_plsda$choice.keepX[1:2]
sel_keepX
## comp1 comp2 
##    20     5
res_splsda <- splsda(X=clean_log_LBA,Y=group_virus_LBA, 
                     ncomp = nlevels(group_virus_LBA),
                     keepX = sel_keepX)
## Warning in cor(A[[k]], variates.A[[k]]): l'écart type est nulle
plotIndiv(res_splsda, comp = c(1,2), ind.names = FALSE, ellipse = TRUE, 
          legend = TRUE, title = 'sPLS-DA (RSV,PI.3 et Coronavirus) - LBA')

head(selectVar(res_splsda, comp = 1)$name)
## [1] "Psychrobacter pulmonis"        "Curtobacterium flaccumfaciens"
## [3] "Comamonas"                     "Paracoccus alcaliphilus"      
## [5] "Brachybacterium sp."           "Moraxella bovoculi 237"
plotLoadings(res_splsda, comp = 1, method = 'mean', contrib = 'max',
             size.title = 1)

plotLoadings(res_splsda, comp = 2, method = 'mean', contrib = 'max',
             size.title = 1)

PLS-DA on PI.3 - EN condition

set.seed(11)
res_plsda <- plsda(X=clean_log_EN,Y=Y_EN$Ct.PI.3, ncomp = nlevels(Y_EN$Ct.PI.3))
## Warning in cor(A[[k]], variates.A[[k]]): l'écart type est nulle
res_perf <- perf(res_plsda, validation = 'Mfold', folds = 5,
                 progressBar = FALSE, nrepeat = 20)
## Warning in MCVfold.splsda(X, Y, multilevel = multilevel, validation = validation, : At least one class is not represented in one fold, which may unbalance the error rate.
##   Consider a number of folds lower than the minimum in table(Y): 2
## Warning in MCVfold.splsda(X, Y, multilevel = multilevel, validation = validation, : At least one class is not represented in one fold, which may unbalance the error rate.
##   Consider a number of folds lower than the minimum in table(Y): 2
plot(res_perf, overlay = 'measure', sd = TRUE)

plotIndiv(res_plsda , comp = c(1, 2), ind.names = FALSE, ellipse = TRUE, 
          legend = TRUE, title = 'PLS-DA (PI.3) - EN')

set.seed(33)
res_plsda <- tune.splsda(X=clean_log_EN,Y=Y_EN$Ct.PI.3, 
                         ncomp = nlevels(Y_EN$Ct.PI.3),
                         test.keepX = 1:20, validation = 'Mfold', 
                         folds = 10, dist = 'mahalanobis.dist', nrepeat = 10,
                         progressBar = F)
## Warning in MCVfold.splsda(X, Y, multilevel = multilevel, validation = validation, : At least one class is not represented in one fold, which may unbalance the error rate.
##   Consider a number of folds lower than the minimum in table(Y): 2

## Warning in MCVfold.splsda(X, Y, multilevel = multilevel, validation = validation, : At least one class is not represented in one fold, which may unbalance the error rate.
##   Consider a number of folds lower than the minimum in table(Y): 2
## Warning: The SGCCA algorithm did not converge

## Warning: The SGCCA algorithm did not converge
plot(res_plsda)

sel_keepX <- res_plsda$choice.keepX[1:2]
sel_keepX
## comp1 comp2 
##    11    11
res_splsda <- splsda(X=clean_log_EN,Y=Y_EN$Ct.PI.3, 
                     ncomp = nlevels(Y_EN$Ct.PI.3),
                     keepX = sel_keepX)
## Warning in cor(A[[k]], variates.A[[k]]): l'écart type est nulle
plotIndiv(res_splsda, comp = c(1,2), ind.names = FALSE, ellipse = TRUE, 
          legend = TRUE, title = 'sPLS-DA (PI.3) - EN')

head(selectVar(res_splsda, comp = 1)$name)
## [1] "Family XIII"              "Parvimonas"              
## [3] "Peptoniphilus"            "Corynebacterium falsenii"
## [5] "Brevundimonas intermedia" "Plantibacter agrosticola"
plotLoadings(res_splsda, comp = 1, method = 'mean', contrib = 'max',
             size.title = 1)

plotLoadings(res_splsda, comp = 2, method = 'mean', contrib = 'max',
             size.title = 1)

PLS-DA on PI.3 - LBA condition

set.seed(11)
res_plsda <- plsda(X=clean_log_LBA,Y=Y_LBA$Ct.PI.3, ncomp = nlevels(Y_LBA$Ct.PI.3))
res_perf <- perf(res_plsda, validation = 'Mfold', folds = 5,
                 progressBar = FALSE, nrepeat = 20)
plot(res_perf, overlay = 'measure', sd = TRUE)
plotIndiv(res_plsda , comp = c(1, 2), ind.names = FALSE, ellipse = TRUE, 
          legend = TRUE, title = 'PLS-DA (PI.3) - LBA')

Error because there is unbalanced class of 1

table(Y_LBA$Ct.PI.3)
## 
##  0  1 
## 21  1
set.seed(33)
res_plsda <- tune.splsda(X=clean_log_LBA,Y=Y_LBA$Ct.PI.3, 
                         ncomp = nlevels(Y_LBA$Ct.PI.3),
                         test.keepX = 1:20, validation = 'Mfold', 
                         folds = 10, dist = 'mahalanobis.dist', nrepeat = 10,
                         progressBar = F)

plot(res_plsda)

sel_keepX <- res_plsda$choice.keepX[1:2]
sel_keepX

Error because there is unbalanced class of 1

res_splsda <- splsda(X=clean_log_LBA,Y=Y_LBA$Ct.PI.3, 
                     ncomp = nlevels(Y_LBA$Ct.PI.3),
                     keepX = sel_keepX)

plotIndiv(res_splsda, comp = c(1,2), ind.names = FALSE, ellipse = TRUE, 
          legend = TRUE, title = 'sPLS-DA (PI.3) - LBA')
head(selectVar(res_splsda, comp = 1)$name)
plotLoadings(res_splsda, comp = 1, method = 'mean', contrib = 'max',
             size.title = 1)
plotLoadings(res_splsda, comp = 2, method = 'mean', contrib = 'max',
             size.title = 1)

Session information

sessionInfo()
## R version 3.4.1 (2017-06-30)
## Platform: x86_64-pc-linux-gnu (64-bit)
## Running under: Ubuntu 16.04.3 LTS
## 
## Matrix products: default
## BLAS: /usr/lib/libblas/libblas.so.3.6.0
## LAPACK: /usr/lib/lapack/liblapack.so.3.6.0
## 
## locale:
##  [1] LC_CTYPE=fr_FR.UTF-8       LC_NUMERIC=C              
##  [3] LC_TIME=fr_FR.UTF-8        LC_COLLATE=fr_FR.UTF-8    
##  [5] LC_MONETARY=fr_FR.UTF-8    LC_MESSAGES=fr_FR.UTF-8   
##  [7] LC_PAPER=fr_FR.UTF-8       LC_NAME=C                 
##  [9] LC_ADDRESS=C               LC_TELEPHONE=C            
## [11] LC_MEASUREMENT=fr_FR.UTF-8 LC_IDENTIFICATION=C       
## 
## attached base packages:
## [1] parallel  stats     graphics  grDevices utils     datasets  methods  
## [8] base     
## 
## other attached packages:
##  [1] randomForest_4.6-12  reshape2_1.4.2       metagenomeSeq_1.18.0
##  [4] RColorBrewer_1.1-2   glmnet_2.0-13        foreach_1.4.3       
##  [7] Matrix_1.2-11        limma_3.32.10        Biobase_2.36.2      
## [10] BiocGenerics_0.22.1  mixOmics_6.3.0       ggplot2_2.2.1       
## [13] lattice_0.20-35      MASS_7.3-47         
## 
## loaded via a namespace (and not attached):
##  [1] gtools_3.5.0       purrr_0.2.4        colorspace_1.2-4  
##  [4] htmltools_0.3.6    yaml_2.1.14        rlang_0.1.4       
##  [7] glue_1.2.0         bindrcpp_0.2       matrixStats_0.52.2
## [10] plyr_1.8.4         bindr_0.1          stringr_1.2.0     
## [13] munsell_0.4.2      gtable_0.1.2       caTools_1.17.1    
## [16] htmlwidgets_0.9    codetools_0.2-15   evaluate_0.10.1   
## [19] labeling_0.3       knitr_1.17         httpuv_1.3.5      
## [22] rARPACK_0.11-0     Rcpp_0.12.13       KernSmooth_2.23-15
## [25] xtable_1.8-2       corpcor_1.6.9      scales_0.5.0      
## [28] backports_1.1.1    gdata_2.18.0       jsonlite_1.5      
## [31] mime_0.5           RSpectra_0.12-0    gplots_3.0.1      
## [34] gridExtra_2.3      ellipse_0.3-8      digest_0.6.12     
## [37] stringi_1.1.5      dplyr_0.7.4        shiny_1.0.5       
## [40] grid_3.4.1         rprojroot_1.2      bitops_1.0-6      
## [43] tools_3.4.1        magrittr_1.5       rgl_0.98.1        
## [46] lazyeval_0.2.1     tibble_1.3.4       tidyr_0.7.2       
## [49] pkgconfig_2.0.1    assertthat_0.2.0   rmarkdown_1.7     
## [52] iterators_1.0.8    R6_2.2.2           igraph_1.1.2      
## [55] compiler_3.4.1